Microvolt amplifier at low-frequency
This discussion is about how certain problems were overcome in the development of a microvolt amplifier.
An application to monitor changes in the range of 0.5 - 5Hz in a very weak magnetic field,
required an amplifier to receive a signal with 3uVrms full-scale to output 1Vrms full-scale with less than 10% noise content.
The input signal comes from a sensor which is, seen electrically, a resistor bridge with 4 x 2k resistors that supplies a differential signal.
The circuit used first comes from conventional design practice and looked as follows:
There is a follower stage to get the signal to 1Vrms but is not shown as it has little relevance.
The total gain of this circuit is estimated to be about 500 000.
As there is only a single rail 5V power supply, <Vref> = 2V was taken a little below ½ rail voltage
to make provision for the op amp saturation when the output is high.
The biggest problem was noise which included large spurious peaks every few seconds.
With A-B short circuited to eliminate sensor noise the following noise sources were checked:
- the op amp
- noise on <Vref>
- surface leakage currents
- resistors and ceramic capacitors
The noise was found to be about 0.5Vrms at the output of the above circuit.
This was lowered to about 0.3Vrms by applying a protective coating on the PCB.
A noise voltage of about 0.15mVpp was found on <Vref> which would seem to account for a large part of the noise on the output.
It was decided to have the first two amplifier stages purely differential to eliminate the noise from any reference voltage. This resulted in the following circuit:
Comparing <Vref> of the 1st circuit with a separate voltage regulator, a zener, 3x1N4148 diodes in series and a green LED, the latter was found to have the lowest noise voltage.
The total gain of this circuit is estimated to be about 2 000 000
and now <Vref> is only introduced at the 3rd stage.
It is desirable to have the largest part of the overall gain in the first stage because then all measures to reduce noise (eg. the quality of the op amp) can be focused on the 1st stage
while the subsequent amplifier stages will then have a small effect on the noise voltage.
The 1st stage gain is however limited by the DC offset voltage of the sensor bridge as well as the op amp offset, so these figures determine the 1st stage gain.
Using the LM324, the noise voltage at the output was about 0.7Vrms
which, adjusting for the differences in gain, corresponds to about 0.18Vrms in the previous circuit. Compared to the 0.5Vrms found in the previous circuit indicates that most of the noise came from <Vref> which was amplified by the 2nd stage.
The op amps in the first 2 stages were now replaced by a low-noise op amp
and the MC33078 was chosen because of its immediate availability at the time.
Instead of 0.7Vrms, the noise voltage at the output was 0.26Vrms.
Having arrived at a relatively low noise level, the effect of resistors and capacitors was checked.
Changes in resistor values and using polyester capacitors instead of ceramics had no measurable effect on the noise voltage.
It has been found however, that the prototype, which is a narrow PCB (15 x 65mm) was very sensitive to mechanical tension in the form of twisting the board.
There are suggestions that at the microvolt levels an op amp can be sensitive to localized temperature variations. This can be verified by heating a solder joint in the vicinity of the 1st stage but is has not been found to be a problem in normal operation.
At this level the noise voltage as measured with the sensor bridge shorted, was found to be a little below the unpredictable voltage variations from the sensor.
There would be an advantage to reduce the noise voltage further and it has subsequently been suggested that in this application the frequency response could be narrowed to 0.5 – 2Hz. This would further reduce the overall noise voltage.
A further measure for this aim would be to find an op amp with even lower noise generation.